Kakao unveils Kanana Space as KakaoTalk expands into AI interface
Summary
Kakao showcased Kanana Space at its annual if(Kakao)26 conference in Yongin, Gyeonggi Province, on Tuesday, outlining plans to make KakaoTalk an interface for messages, schedules and shopping. The assistant will begin a closed beta test in December and use voice, text and conversational context to complete tasks. Kakao also introduced Kanvas for building teams of AI agents and plans to add an agent-access space to KakaoTalk in the first half of 2027.
Key Facts
- Kanana Space can be opened anywhere in KakaoTalk by swiping the screen with two fingers and can retrieve information from conversations, organize unread messages, manage schedules and adjust app settings.
- Kakao plans to extend Kanana Space across shopping, gifting, reservations and money transfers, while requiring users to review and approve consequential actions such as sending messages and making payments.
- Kanvas allows users to configure AI agents for specific tasks, preserve context through local memory and run recurring or scheduled tasks on personal computers.
- Kanvas uses separate browsers for each agent to isolate access information and limit actions to the scope authorized by users.
- Kakao is developing a system that selects AI models of different sizes according to request complexity to balance performance, speed and computing costs.
Kakao CEO Chung Shin-a speaks during the company's annual if(Kakao)26 conference at Kakao AI Campus in Yongin, Gyeonggi Province, Tuesday. Courtesy of Kakao
Kakao showcased an artificial intelligence (AI) assistant, Kanana Space, for KakaoTalk on Tuesday, outlining plans to turn its flagship messenger into an interface for managing messages, schedules and shopping across its service ecosystem.
“Over the past 15 years, Kakao has connected people with everyday life and bridged online and offline experiences with the goal of connecting everything in the world,” the company’s CEO Chung Shin-a said during its annual if(Kakao)26 conference at Kakao AI Campus in Yongin, Gyeonggi Province.
“We now aim to take the connections we have built to a new level.”
The service, which is slated to begin a closed beta test in December, can be accessed from anywhere in KakaoTalk by swiping the screen with two fingers. Users can interact with it through voice or text to retrieve information from conversations, organize unread messages, manage schedules and adjust app settings.
Kanana Space is designed to go beyond answering questions by using context from users’ past interactions to complete tasks. In a demonstration, the assistant identified work-related messages, summarized key items and drafted a reply.
The company plans to expand the assistant’s capabilities to handle tasks across its various services, including shopping, gifting, reservations and money transfers.
Kim Dae-nyeon, the company’s head of design, said users would retain control over consequential actions handled by the assistant.
“For important actions such as sending messages or making payments, users must check and give final approval themselves,” Kim said.
Kim Dae-nyeon, Kakao's head of design, demonstrates Kanana Space during the company's annual if(Kakao)26 conference at Kakao AI Campus in Yongin, Gyeonggi Province, Tuesday. Courtesy of Kakao
Kakao also introduced Kanvas, a tool designed to let users build and run their own teams of AI agents on personal computers. The tool allows users to configure agents for specific tasks, retain context through local memory and run recurring or scheduled tasks.
The company also plans to create a new space in KakaoTalk in the first half of 2027, where users can access their own teams of AI agents, including those built by Kakao, its partners and users, as well as agents developed using other frameworks.
“Kakao aims to become a common point of access for AI agents in users’ everyday lives,” Chung said. “We plan to connect agents developed by Kakao, its partners and users through a single, familiar interface.”
Chung said Kanvas uses separate browsers for each agent to keep access information isolated, so that the sandboxed environment limits agents’ actions to the scope authorized by users.
Beyond its agentic AI services, Kakao is developing an AI system that selects models of different sizes based on the complexity of each request, aiming to balance performance, speed and computing costs.
Roh Byung-seok, head of foundation models, said the approach is designed to handle a wide range of requests, from simple text edits to complex reasoning tasks without always relying on a frontier model.
“Our principle is simple: Deliver the best results for every request, but don’t use the largest model for every request,” Roh said.
Explore More
- Q.
- Q.
- Q.